Improved performance of utf8 validation of large strings via simdutf8 (-40%)

Cargo.toml

@@ -64,8 +64,13 @@ avro-rs = { version = "0.13", optional = true, default_features = false }
 
 # for division/remainder optimization at runtime
 strength_reduce = { version = "0.2", optional = true }
+
+# For instruction multiversioning
 multiversion = { version = "0.6.1", optional = true }
 
+# For SIMD utf8 validation
+simdutf8 = "0.1.3"
+
 [dev-dependencies]
 rand = "0.8"
 criterion = "0.3"

benches/read_parquet.rs

@@ -42,6 +42,11 @@ fn add_benchmark(c: &mut Criterion) {
             b.iter(|| read_decompressed_pages(&buffer, size * 8, 2).unwrap())
         });
 
+        let a = format!("read utf8 large 2^{}", i);
+        c.bench_function(&a, |b| {
+            b.iter(|| read_decompressed_pages(&buffer, size * 8, 6).unwrap())
+        });
+
         let a = format!("read bool 2^{}", i);
         c.bench_function(&a, |b| {
             b.iter(|| read_decompressed_pages(&buffer, size * 8, 3).unwrap())

parquet_integration/write_parquet.py

@@ -10,6 +10,7 @@ def case_basic_nullable(size=1):
     float64 = [0.0, 1.0, None, 3.0, None, 5.0, 6.0, 7.0, None, 9.0]
     string = ["Hello", None, "aa", "", None, "abc", None, None, "def", "aaa"]
     boolean = [True, None, False, False, None, True, None, None, True, True]
+    string_large = ["ABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCD😃🌚🕳👊"] * 10
 
     fields = [
         pa.field("int64", pa.int64()),
@@ -18,6 +19,7 @@ def case_basic_nullable(size=1):
         pa.field("bool", pa.bool_()),
         pa.field("date", pa.timestamp("ms")),
         pa.field("uint32", pa.uint32()),
+        pa.field("string_large", pa.utf8()),
     ]
     schema = pa.schema(fields)
 
@@ -29,6 +31,7 @@ def case_basic_nullable(size=1):
             "bool": boolean * size,
             "date": int64 * size,
             "uint32": int64 * size,
+            "string_large": string_large * size,
         },
         schema,
         f"basic_nullable_{size*10}.parquet",

src/array/ord.rs

@@ -48,51 +48,90 @@ where
 }
 
 fn compare_primitives<T: NativeType + Ord>(left: &dyn Array, right: &dyn Array) -> DynComparator {
-    let left = left.as_any().downcast_ref::<PrimitiveArray<T>>().unwrap().clone();
-    let right = right.as_any().downcast_ref::<PrimitiveArray<T>>().unwrap().clone();
+    let left = left
+        .as_any()
+        .downcast_ref::<PrimitiveArray<T>>()
+        .unwrap()
+        .clone();
+    let right = right
+        .as_any()
+        .downcast_ref::<PrimitiveArray<T>>()
+        .unwrap()
+        .clone();
     Box::new(move |i, j| total_cmp(&left.value(i), &right.value(j)))
 }
 
 fn compare_boolean(left: &dyn Array, right: &dyn Array) -> DynComparator {
-    let left = left.as_any().downcast_ref::<BooleanArray>().unwrap().clone();
-    let right = right.as_any().downcast_ref::<BooleanArray>().unwrap().clone();
+    let left = left
+        .as_any()
+        .downcast_ref::<BooleanArray>()
+        .unwrap()
+        .clone();
+    let right = right
+        .as_any()
+        .downcast_ref::<BooleanArray>()
+        .unwrap()
+        .clone();
     Box::new(move |i, j| left.value(i).cmp(&right.value(j)))
 }
 
 fn compare_f32(left: &dyn Array, right: &dyn Array) -> DynComparator {
-    let left = left.as_any().downcast_ref::<PrimitiveArray<f32>>().unwrap().clone();
+    let left = left
+        .as_any()
+        .downcast_ref::<PrimitiveArray<f32>>()
+        .unwrap()
+        .clone();
     let right = right
         .as_any()
         .downcast_ref::<PrimitiveArray<f32>>()
-        .unwrap().clone();
+        .unwrap()
+        .clone();
     Box::new(move |i, j| total_cmp_f32(&left.value(i), &right.value(j)))
 }
 
 fn compare_f64(left: &dyn Array, right: &dyn Array) -> DynComparator {
-    let left = left.as_any().downcast_ref::<PrimitiveArray<f64>>().unwrap().clone();
+    let left = left
+        .as_any()
+        .downcast_ref::<PrimitiveArray<f64>>()
+        .unwrap()
+        .clone();
     let right = right
         .as_any()
         .downcast_ref::<PrimitiveArray<f64>>()
-        .unwrap().clone();
+        .unwrap()
+        .clone();
     Box::new(move |i, j| total_cmp_f64(&left.value(i), &right.value(j)))
 }
 
 fn compare_string<O: Offset>(left: &dyn Array, right: &dyn Array) -> DynComparator {
-    let left = left.as_any().downcast_ref::<Utf8Array<O>>().unwrap().clone();
-    let right = right.as_any().downcast_ref::<Utf8Array<O>>().unwrap().clone();
+    let left = left
+        .as_any()
+        .downcast_ref::<Utf8Array<O>>()
+        .unwrap()
+        .clone();
+    let right = right
+        .as_any()
+        .downcast_ref::<Utf8Array<O>>()
+        .unwrap()
+        .clone();
     Box::new(move |i, j| left.value(i).cmp(right.value(j)))
 }
 
 fn compare_binary<O: Offset>(left: &dyn Array, right: &dyn Array) -> DynComparator {
-    let left = left.as_any().downcast_ref::<BinaryArray<O>>().unwrap().clone();
-    let right = right.as_any().downcast_ref::<BinaryArray<O>>().unwrap().clone();
+    let left = left
+        .as_any()
+        .downcast_ref::<BinaryArray<O>>()
+        .unwrap()
+        .clone();
+    let right = right
+        .as_any()
+        .downcast_ref::<BinaryArray<O>>()
+        .unwrap()
+        .clone();
     Box::new(move |i, j| left.value(i).cmp(right.value(j)))
 }
 
-fn compare_dict<K>(
-    left: &DictionaryArray<K>,
-    right: &DictionaryArray<K>,
-) -> Result<DynComparator>
+fn compare_dict<K>(left: &DictionaryArray<K>, right: &DictionaryArray<K>) -> Result<DynComparator>
 where
     K: DictionaryKey,
 {

src/array/specification.rs

@@ -79,7 +79,8 @@ pub fn check_offsets_and_utf8<O: Offset>(offsets: &[O], values: &[u8]) -> usize
         let end = window[1].to_usize();
         assert!(end <= values.len());
         let slice = unsafe { std::slice::from_raw_parts(values.as_ptr().add(start), end - start) };
-        std::str::from_utf8(slice).expect("A non-utf8 string was passed.");
+        #[cfg(not(feature = "simdutf8"))]
+        simdutf8::basic::from_utf8(slice).expect("A non-utf8 string was passed.");
     });
     len
 }

src/compute/cast/mod.rs

@@ -538,7 +538,7 @@ fn cast_with_options(
                 // perf todo: the offsets are equal; we can speed-up this
                 let iter = array
                     .iter()
-                    .map(|x| x.and_then(|x| std::str::from_utf8(x).ok()));
+                    .map(|x| x.and_then(|x| simdutf8::basic::from_utf8(x).ok()));
 
                 let array = Utf8Array::<i32>::from_trusted_len_iter(iter);
                 Ok(Box::new(array))
@@ -574,7 +574,7 @@ fn cast_with_options(
                 // perf todo: the offsets are equal; we can speed-up this
                 let iter = array
                     .iter()
-                    .map(|x| x.and_then(|x| std::str::from_utf8(x).ok()));
+                    .map(|x| x.and_then(|x| simdutf8::basic::from_utf8(x).ok()));
 
                 let array = Utf8Array::<i64>::from_trusted_len_iter(iter);
                 Ok(Box::new(array))

src/compute/like.rs

@@ -173,7 +173,7 @@ fn a_like_binary<O: Offset, F: Fn(bool) -> bool>(
                     let pattern = if let Some(pattern) = map.get(pattern) {
                         pattern
                     } else {
-                        let re_pattern = std::str::from_utf8(pattern)
+                        let re_pattern = simdutf8::basic::from_utf8(pattern)
                             .unwrap()
                             .replace("%", ".*")
                             .replace("_", ".");
@@ -231,7 +231,7 @@ fn a_like_binary_scalar<O: Offset, F: Fn(bool) -> bool>(
     op: F,
 ) -> Result<BooleanArray> {
     let validity = lhs.validity();
-    let pattern = std::str::from_utf8(rhs).map_err(|e| {
+    let pattern = simdutf8::basic::from_utf8(rhs).map_err(|e| {
         ArrowError::InvalidArgumentError(format!(
             "Unable to convert the LIKE pattern to string: {}",
             e

src/error.rs

@@ -51,6 +51,12 @@ impl From<std::str::Utf8Error> for ArrowError {
     }
 }
 
+impl From<simdutf8::basic::Utf8Error> for ArrowError {
+    fn from(error: simdutf8::basic::Utf8Error) -> Self {
+        ArrowError::External("".to_string(), Box::new(error))
+    }
+}
+
 impl Display for ArrowError {
     fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
         match self {

src/ffi/schema.rs

@@ -435,7 +435,7 @@ unsafe fn read_ne_i32(ptr: *const u8) -> i32 {
 
 unsafe fn read_bytes(ptr: *const u8, len: usize) -> &'static str {
     let slice = std::slice::from_raw_parts(ptr, len);
-    std::str::from_utf8(slice).unwrap()
+    simdutf8::basic::from_utf8(slice).unwrap()
 }
 
 unsafe fn metadata_from_bytes(data: *const ::std::os::raw::c_char) -> (Metadata, Extension) {

src/io/avro/read/deserialize.rs

@@ -104,7 +104,7 @@ pub fn deserialize(mut block: &[u8], rows: usize, schema: Arc<Schema>) -> Result
                             "Avro format contains a non-usize number of bytes".to_string(),
                         )
                     })?;
-                    let data = std::str::from_utf8(&block[..len])?;
+                    let data = simdutf8::basic::from_utf8(&block[..len])?;
                     block = &block[len..];
 
                     let array = array

src/io/avro/read/util.rs

@@ -74,7 +74,7 @@ pub fn read_schema<R: Read>(reader: &mut R) -> AvroResult<(Schema, Codec, [u8; 1
             .get("avro.codec")
             .and_then(|codec| {
                 if let Value::Bytes(ref bytes) = *codec {
-                    std::str::from_utf8(bytes.as_ref()).ok()
+                    simdutf8::basic::from_utf8(bytes.as_ref()).ok()
                 } else {
                     None
                 }

src/io/csv/read/deserialize.rs

@@ -52,7 +52,7 @@ where
 
 fn deserialize_utf8<O: Offset>(rows: &[ByteRecord], column: usize) -> Arc<dyn Array> {
     let iter = rows.iter().map(|row| match row.get(column) {
-        Some(bytes) => std::str::from_utf8(bytes).ok(),
+        Some(bytes) => simdutf8::basic::from_utf8(bytes).ok(),
         None => None,
     });
     Arc::new(Utf8Array::<O>::from_trusted_len_iter(iter))
@@ -111,44 +111,44 @@ pub fn deserialize_column(
             lexical_core::parse::<f64>(bytes).ok()
         }),
         Date32 => deserialize_primitive(rows, column, datatype, |bytes| {
-            std::str::from_utf8(bytes)
+            simdutf8::basic::from_utf8(bytes)
                 .ok()
                 .and_then(|x| x.parse::<chrono::NaiveDate>().ok())
                 .map(|x| x.num_days_from_ce() - EPOCH_DAYS_FROM_CE)
         }),
         Date64 => deserialize_primitive(rows, column, datatype, |bytes| {
-            std::str::from_utf8(bytes)
+            simdutf8::basic::from_utf8(bytes)
                 .ok()
                 .and_then(|x| x.parse::<chrono::NaiveDateTime>().ok())
                 .map(|x| x.timestamp_millis())
         }),
         Timestamp(TimeUnit::Nanosecond, None) => {
             deserialize_primitive(rows, column, datatype, |bytes| {
-                std::str::from_utf8(bytes)
+                simdutf8::basic::from_utf8(bytes)
                     .ok()
                     .and_then(|x| x.parse::<chrono::NaiveDateTime>().ok())
                     .map(|x| x.timestamp_nanos())
             })
         }
         Timestamp(TimeUnit::Microsecond, None) => {
             deserialize_primitive(rows, column, datatype, |bytes| {
-                std::str::from_utf8(bytes)
+                simdutf8::basic::from_utf8(bytes)
                     .ok()
                     .and_then(|x| x.parse::<chrono::NaiveDateTime>().ok())
                     .map(|x| x.timestamp_nanos() / 1000)
             })
         }
         Timestamp(TimeUnit::Millisecond, None) => {
             deserialize_primitive(rows, column, datatype, |bytes| {
-                std::str::from_utf8(bytes)
+                simdutf8::basic::from_utf8(bytes)
                     .ok()
                     .and_then(|x| x.parse::<chrono::NaiveDateTime>().ok())
                     .map(|x| x.timestamp_nanos() / 1_000_000)
             })
         }
         Timestamp(TimeUnit::Second, None) => {
             deserialize_primitive(rows, column, datatype, |bytes| {
-                std::str::from_utf8(bytes)
+                simdutf8::basic::from_utf8(bytes)
                     .ok()
                     .and_then(|x| x.parse::<chrono::NaiveDateTime>().ok())
                     .map(|x| x.timestamp_nanos() / 1_000_000_000)

src/io/parquet/read/schema/metadata.rs

@@ -113,7 +113,7 @@ mod tests {
             .collect::<Vec<_>>();
         assert_eq!(
             names,
-            vec!["int64", "float64", "string", "bool", "date", "uint32"]
+            vec!["int64", "float64", "string", "bool", "date", "uint32", "string_large"]
         );
         Ok(())
     }

src/io/parquet/read/statistics/binary.rs

@@ -57,12 +57,12 @@ impl TryFrom<&ParquetByteArrayStatistics> for Utf8Statistics {
             min_value: stats
                 .min_value
                 .as_ref()
-                .map(|x| std::str::from_utf8(x).map(|x| x.to_string()))
+                .map(|x| simdutf8::basic::from_utf8(x).map(|x| x.to_string()))
                 .transpose()?,
             max_value: stats
                 .max_value
                 .as_ref()
-                .map(|x| std::str::from_utf8(x).map(|x| x.to_string()))
+                .map(|x| simdutf8::basic::from_utf8(x).map(|x| x.to_string()))
                 .transpose()?,
         })
     }